In this paper, we address the problem of dynamic network embedding, that is, representing the nodes of a dynamic network as evolving vectors within a low-dimensional space. While the field of single-graph embedding is wide and established, the field of dynamic graph embedding is comparatively in its infancy. In this paper, we propose that we can take a wide class of established single-graph embedding methods and use them to produce interpretable and powerful dynamic graph embeddings by simply applying them to the dilated unfolded matrix. We provide a theoretical guarantee that, regardless of embedding dimension, these unfolded methods will produce stable embeddings over time and space, meaning that nodes with identical latent behaviour will be exchangeable, regardless of their position in time or space.
Contact & Links
Email: edward.davis[at]bristol.ac.uk
Office: School of Mathematics University of Bristol, Fry Building, GA.06 Woodland Road Bristol, BS8 1UG.
Twitter: EdD8ta
LinkedIn: edwarddavis941
GitHub: edwarddavis1
PhD supervised by Prof. Patrick Rubin-Delanchy and Dr Daniel Lawson.
Research Outline
My work focuses on dynamic network embedding. Dynamic networks encode both spatial and temporal relations between nodes, which are desirable for many applications from cyber security to neuroscience. By representing the nodes of the network as a vector space, it allows us to detect spatial and temporal patterns through clustering. Methods that I have developed during my PhD have been used by Microsoft and LV, and I have presented these methods at the Institute of Mathematics Big Data Conference and the COMPASS conference in 2022. I am interested in progressing dynamic embedding methods and applying them to a diverse range of problems.
Papers
A Simple but Powerful Framework for Dynamic Graph Embedding
Valid Conformal Prediction for Dynamic GNNs
Valid Bootstraps for Networks with Applications to Network Visualisation
Quantifying uncertainty in networks is an important step in modelling relationships and interactions between entities. We consider the challenge of bootstrapping an inhomogeneous random graph when only a single observation of the network is made and the underlying data generating function is unknown. We utilise an exchangeable network test that can empirically validate bootstrap samples generated by any method, by testing if the observed and bootstrapped networks are statistically distinguishable. We find that existing methods fail this test. To address this, we propose a principled, novel, distribution-free network bootstrap using k-nearest neighbour smoothing, that can regularly pass this exchangeable network test in both synthetic and real-data scenarios. We demonstrate the utility of this work in combination with the popular data visualisation method t-SNE, where uncertainty estimates from bootstrapping are used to explain whether visible structures represent real statistically sound structures.
Accomplishments
Award-Winning Data Visualisation
Runner-up in the Jean Golding Institute's Beauty of Data competition. This visualisation was presented at the Bristol Data and AI Showcase 2022. This visualisation displays an embedding of a friendship network between countries on the world stage based on their alliances.
(Click image for full-size)